Homework Assignment Computational Argumentation 2026/2027

The homework assignment concerns the modelling of a realistic argumentation example in the TOAST implementation for ASPIC+. The exercise must be done in pairs. Have a debate between the two of you about a claim. (You can use Generative AI in any way you like to help you generating your debate.) Then model the arguments that you exchanged in your debate by specifying a TOAST input. (You can use Generative AI in any way you like to help you generating your Toast modelling.) Use the category of 'premises' for ordinary premises (Kp) and the category of 'axioms' for necessary premises (Kn). Leave the category of assumption premises empty but provide rule and/or knowledge-base preferences if possible. Your modelling should contain at least 10 rules and generate at least five separate arguments (no subargument relations between them) with at least three symmetric or asymmetric defeat relations between them.

Hand in your modelling and a report by email to Henry Prakken as a single document containing the following information:

• A natural-language version of your debate;
• a URL of your TOAST modelling and (if you have used rule preferences) whether you have chosen the weakest- or last-link argument ordering;
• a graphical visualisation of your modelling as in e.g. Figure 4.5 of the reader (among other things showing the TOAST rule names where they are used);
• a description of max 2 pages of your modelling method, including an explanation of your design and modelling choices and discussions of the following points:
  • Whether you have used argument schemes for modelling the arguments.
  • What you think of the outcome of the debate and how it depends on the chosen semantics.
  • Whether limitations of the software (if any) caused you any problems.
• If you used generative AI to generate your debate or modelling: a trace of your conversation(s) with the generative-AI tool.

Assessment criteria:

Your assignment will be assessed on three aspects, which all count equally: (1) correctness of your modelling, (2) elegance and substance of your modelling, and (3) correctness and substance of your report. You can be asked to give an oral explanation of your modelling and report as part of the assessment.

Software:

• Toast, an implementation of an ASPIC+ instantiation developed by Arg-Tech, at Dundee (Scotland). Click on "help" for information on syntax and use of the system. Note that contrary to what the TOAST help file says, there must not be a space before or after a rule's implication. The visualiser of Dung frameworks is slightly buggy: it does not always refresh earlier shown graphs.
  
  Note also that TOAST deviates from the APSIC+ theory as described in the reader in a few minor ways:
  • TOAST does not allow rules with empty antecedents.
  • NB: Toast does not allow the formulation of undercutters as premises. So, for instance, with a defeasible rule [r1] p => q and a premise ~[r1}, Toast gives the warning "the argumentation theory is not well-formed."
  • If p and q are ordinary premises and TOAST has a defeasible rule q => -p, then Toast does not recognise that LastDefRules(p) is empty while LastDefRules(q => -p) is nonempty so that p should, according to the reader, with the last-link ordering still be preferred (and also in the weakest-link ordering if p is preferred over q). Apparently, Toast implements slightly different versions of these orderings.